Wednesday, March 23, 2016

pyeviews: Python + EViews


Since we love Python (who doesn’t?), we’ve had it in the back of our minds for a while now that we should find a way to make it easier for EViews and Python to talk to each other, so Python programmers can use the econometric engine of EViews directly from Python. So we did! We’ve written a Python package called pyeviews that uses COM to transfer data between Python and EViews (For more information on COM and EViews, take a look at our whitepaper on the subject).

Here’s a simple example going from Python to EViews. We’re going to use the popular Chow-Lin interpolation routine in EViews using data created in Python. Chow-Lin interpolation is a regression-based technique to transform low-frequency data (in our example, annual) into higher-frequency data (in our example, quarterly). It has the ability to use a higher-frequency series as a pattern for the interpolated series to follow. The quarterly interpolated series is chosen to match the annual benchmark series in one of four ways: first (the first quarter value of the interpolated series matches the annual series), last (same, but for the fourth quarter value), sum (the sum of the first through fourth quarters matches the annual series), and average (the average of the first through fourth quarters matches the annual series).

Friday, March 18, 2016

How We Decide Which Features To Add


As developers of econometric software, one of the most common questions we are asked is how we decide which features to add to the next release of EViews.

There isn’t an easy way to answer this question – the process is often fluid and is different for every feature. Feature ideas generally come to us from one of the following sources:
·
  • Directly from our user base, either on the EViews forum, through our technical support channels or through face to face meetings. 
  • From reading journal articles and text books to discover the latest trends in the field.
  • From visiting academic and professional conferences, such as ASSA, NABE or ISF.
  • From meetings held at our user conferences.
  • Research from our development team.

The recent release of EViews 9.5 gives us a chance to explore the process with some examples.

MIDAS

Perhaps the most anticipated feature in EViews 9.5 is MIDAS estimation, which allows estimation of regression models using data of different frequencies. MIDAS first came to our attention a few years ago during a casual conversation between our developers and one of our academic users at the Joint Statistical Meetings.

Our user suggested that EViews’ natural handling of data from different frequencies and our emphasis on time series analysis, coupled with MIDAS’ growing popularity made it a great candidate for a new feature.

Following up on that discussion, our development team began researching what would be involved in adding MIDAS to EViews. MIDAS would be the first estimation technique in EViews that inherently uses data based on different workfile pages.

While later attending the EViews co-sponsored ISF conference, we also noticed that a large part of the conference was devoted to MIDAS estimation and forecasting as well as nowcasting. By this time we were convinced that MIDAS was an obvious choice to add to EViews.